A nonprofit program lead plans to use an agent to screen volunteer applications and summarize qualifications for coordinators. The data includes names and contact details, and decisions affect opportunities fairly. Which application of Responsible AI principles to agent use best guides this deployment?

The correct approach aligns with Responsible AI across the full screening workflow because volunteer applications contain personal identifiers requiring data minimization and purpose-limited access, affected individuals deserve transparency about AI assistance in summaries, fairness-sensitive opportunity decisions require meaningful human accountability rather than autonomous action, and summarization plus ranking risks demand ongoing evaluation for accuracy and bias. The broad-access variant fails on privacy and least privilege by trading minimization for convenience even while preserving disclosure, human review, and fairness monitoring, which still violates stewardship of names and contact details. The validation-only variant fails on reliability and fairness maintenance because pre-launch testing cannot catch drift, changing applicant pools, or emergent errors, so continuous monitoring is essential where decisions affect opportunities fairly. The nondisclosure variant fails on transparency because coordinators acting on AI summaries need to know assistance occurred to calibrate trust, verify details, and maintain accountability, even when access limits, human decisions, and monitoring are strong. On exams, avoid selecting an answer that is operationally sound but incomplete when the scenario explicitly names privacy, fairness, and human impact together. As a practical check, confirm the agent sees only fields needed for the summary, logs AI involvement visible to coordinators, blocks final action without human approval, and schedules periodic accuracy and disparity reviews with remediation.